Yes Igor. You are correct. We should pass the info we have stored in the serialization so that they can initialize correctly :) On Thu, Jun 2, 2011 at 3:42 PM, Igor Stasenko <siguctua@gmail.com> wrote:
On 2 June 2011 15:22, Mariano Martinez Peck <marianopeck@gmail.com> wrote:
On Thu, Jun 2, 2011 at 3:00 PM, Igor Stasenko <siguctua@gmail.com>
wrote:
Btw, what deserializer does, if it discovers that instance of some class, which you want to bring to life from storage now has its format changed? suppose that when saving graph, some class has inst vars 'a b c' but then you changed the class definition and its inst vars now 'a d e' ?
In this, Fuel will materialize the object, set the instVar 'a' and 'd' and 'e' will be with nil. After, the user of FUel can do whatever he wants. In fact, we will provide soemthing like #initializeFromFuel or something like that that is automatically called by Fuel after the materialization. We throw error where there is something we cannot do.
Well, then it should be initializeFromFuel: fuelSpec
instead of just initializeFromFuel
because if you don't know, what were stored in fuel stream, you cannot properly migrate the instance. Consider that initially class has inst vars 'a' and you saved this instance to fuel. Then you changed ivars to 'a b' and changed the #initializeFromFuel appropriately. But then you again changed ivars to 'c b d' and also changed #initializeFromFuel Now if you attempt to materialize the instance from first version (with only 'a'), you cannot determine what original format of instance were stored in fuel, and so, you cannot decide what will be a proper migration path: 'a' -> 'c b d' or 'a b' -> 'c b d' or 'c b d' -> 'c b d'
and i think it is better to move this method to class side, then it can deal with migration before creating an instance. So you can even replace an instance of original class with something else (if you refactored your model and no longer want an instances of old class, so you stub a creation of them in class side of old class, and create instances of different class instead).
Or even two-phase materialization:
SomeClass class>>materializeFromFuel: fuelSpec ... "some common logic" .. ^ self basicNew initializeFromFuel: fuelSpec
so, you will have two polymorphic entry points.
I think it is not a responsibility of serializer to deal with that (there should be a higher level layer which can handle instance migration), while serializer just reports an error that it can't reify an object because its original class are either gone or changed format.
yes, but there are things that we can do (kind of first pass) so that we can give the user the object already mterialized and he can update it.
In fuel, we have these tests:
testVariableInsertion "Tests that serializer tolarates when there is a new instance variable on materialization"
| stream aPair resultPair |
aPair := (self stubClassWithInstanceVars: 'left right') new. aPair instVarAt: 1 put: $A. aPair instVarAt: 2 put: $B. stream := self serializationOf: aPair.
self stubClassWithInstanceVars: 'left middle right'. resultPair := self materializationOn: stream reset. self assert: $A equals: (resultPair instVarAt: 1). self assert: nil equals: (resultPair instVarAt: 2). self assert: $B equals: (resultPair instVarAt: 3).
testVariableRemoved "Tests that serializer tolarates when an instance variable is missing on materialization"
| stream aPair resultPair |
aPair := (self stubClassWithInstanceVars: 'left right') new. aPair instVarAt: 1 put: $A. aPair instVarAt: 2 put: $B. stream := self serializationOf: aPair.
self stubClassWithInstanceVars: 'right'. resultPair := self materializationOn: stream reset. self assert: $B equals: (resultPair instVarAt: 1).
testVariableOrderChange "Tests that serializer tolarates when the order in the instance variables changed between serialization and materialization"
| pairClass stream aPair resultPair |
pairClass := self stubClassWithInstanceVars: 'left right'. aPair := pairClass new. aPair instVarAt: 1 put: $A. aPair instVarAt: 2 put: $B. stream := self serializationOf: aPair.
pairClass := self stubClassWithInstanceVars: 'right left'. resultPair := self materializationOn: stream reset. self assert: $B equals: (resultPair instVarAt: 1). self assert: $A equals: (resultPair instVarAt: 2).
On 2 June 2011 14:51, Norbert Hartl <norbert@hartl.name> wrote:
Am 02.06.2011 um 14:31 schrieb Mariano Martinez Peck:
Thanks Norbert. Now I could successfully run benchmarks for SIXX and compare with Fuel. But that's not really fare because we are comparing a text based serializer against a binary one.
It is interesting anyway. I like to know if it is 20 times faster or even more. On the other hand it is not fair either. I don't know fuel but I think it platform dependent, right? So it is not fair the other way round because you compare a cross-platform serializer with a platform-dependent one
;)
I think these are the categories that people think about when the are about to choose what util to use. Norbert
Mariano
On Thu, Jun 2, 2011 at 12:14 PM, Norbert Hartl <norbert@hartl.name> wrote:
Am 02.06.2011 um 11:42 schrieb Mariano Martinez Peck:
serialize: anObject on: aStream | sws | sws := SixxWriteStream on: aStream. sws nextPut: anObject. sws close.
serialize: anObject on: aStream anObject sixxOn: aStream
materializeFrom: aStream | srs objects | srs := SixxReadStream on: aStream. objects := srs contents. srs close. ^ objects
materializeFrom: aStream ^ Object readSixxFrom: aStream Norbert
Is that correct or I am doing it wrong? All I want to do is to serialize a graph into a stream.
The stream I am using is or this:
(FileDirectory default forceNewFileNamed: 'Bench')
or
(RWBinaryOrTextStream on: '')
I am not sure if I should be doing a #nextPut: or a #nextPutAll:. I mean, sometimes anObject is a collection and sometimes it is not. So, which one should I use?
Thanks
Mariano
-- Mariano http://marianopeck.wordpress.com
-- Mariano http://marianopeck.wordpress.com
-- Best regards, Igor Stasenko AKA sig.
-- Mariano http://marianopeck.wordpress.com
-- Best regards, Igor Stasenko AKA sig.
-- Mariano http://marianopeck.wordpress.com